Add MiniMax as a first-class LLM provider

[octo-patch]

Summary

Adds MiniMax as a built-in LLM provider, following the same OpenAI-compatible pattern used by DeepSeek and Perplexity.

MiniMax offers high-performance language models with up to 1M token context windows through an OpenAI-compatible API.

Changes

Provider implementation (5 new files)

• providers/minimax.rb — Main provider class extending OpenAI, custom api_base and headers
• providers/minimax/chat.rb — Role formatting
• providers/minimax/capabilities.rb — Model metadata, pricing, and feature flags
• providers/minimax/models.rb — Static model listing (MiniMax has no /v1/models endpoint)
• providers/minimax/temperature.rb — Clamps temperature to [0.0, 1.0] with debug logging

Models supported

Model	Context Window	Notes
MiniMax-M2.7	1,000,000	Latest flagship model
MiniMax-M2.7-highspeed	1,000,000	Cost-optimized variant
MiniMax-M2.5	204,000	Previous generation
MiniMax-M2.5-highspeed	204,000	Cost-optimized variant

Configuration

RubyLLM.configure do |config|
  config.minimax_api_key = ENV['MINIMAX_API_KEY']
  config.minimax_api_base = ENV['MINIMAX_API_BASE'] # optional
end

Tests (5 spec files, 48 examples, 0 failures)

• minimax_spec.rb — Provider initialization, API base, headers, configuration, slug
• minimax/capabilities_spec.rb — Context windows, model families, pricing, features
• minimax/temperature_spec.rb — Clamping, nil handling, edge cases
• minimax/models_spec.rb — Static model list, metadata, provider slug
• minimax/chat_spec.rb — Role formatting

Other changes

• Registered provider in lib/ruby_llm.rb with Zeitwerk inflection
• Added to .env.example, README.md, configuration docs, test support files

Test plan

• All 48 new unit tests pass
• Existing provider_spec.rb tests pass (configuration schema validation)
• CI should pass on all supported Ruby versions